Seasonality and the COVID-19 shock in Greek road accidents (2010-2024): Interrupted time-series evidence and diagnostic forecasts for operational road safety planning.
Abstract
Objective
We analyzed monthly administrative data on road crashes and casualties in Greece for the period April 2010-June 2024 (N = 171), with the aim of supporting operational road safety planning.
Methods
We applied interrupted time series analysis (ITSA) to separate stable seasonal patterns from the impact of the Corona virus disease 2019 (COVID-19) shock. The main analysis is based on segmented regression (ITSA) with monthly dummies for seasonality, estimated using ordinary least squares (OLS) with Newey-West heteroskedasticity and autocorrelation consistent (HAC) errors, with quasi-Poisson models used as count-data robustness checks. Outcomes include monthly counts of crashes, fatalities, and serious and slight injuries.
Results
Road safety indicators peak in June-August and reach their lowest levels in January-March. In March 2020 there was an abrupt level drop in all outcomes (crashes, fatalities, serious injuries, slight injuries) (followed by only partial recovery). The rebound after March 2020 is more visible for total crashes and slight injuries. For fatalities and serious injuries, the count-data checks do not show the same clear upward pattern. This indicates stabilization at a lower post-COVID-19 baseline for the most severe outcomes. Pre/post comparisons around March 2020 indicate large observed differences relative to pre-pandemic average monthly levels, corresponding descriptively to approximately 7,200 fewer crashes and about 1,070 fewer deaths over March 2020-June 2024. We refer to absolute monthly numbers, which are useful for planning system load. They should not be interpreted as exposure-adjusted individual crash risk, because monthly exposure measures such as vehicle-kilometres traveled (VKT) were not available for the full study period.
Conclusions
The study is quasi-experimental and describes associations. It does not establish full causal attribution. All data and code are provided as supplementary files.